Actual Connection Time

8/12/2019 Actual Connection Time

1/10

682 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 1, NO. 4, OCTOBER 2002

Actual Call Connection Time Characterizationfor Wireless Mobile Networks Under a

General Channel Allocation SchemeWei Li, Member, IEEE, Yuguang Fang, Senior Member, IEEE, and Robert R. Henry, Senior Member, IEEE

AbstractActual call connection time (ACCT) is the total timethat a mobile user engages in communications over a wireless net-work during a call connection. Due to limited network resourcesof wireless mobile networks, a call connection may be prematurelydisconnected and the ACCT for the call in general may not be thesame as the requested call connection time (RCCT). The ACCTdepends not only on the RCCT, but also on the network resourceallocation scheme and network traffic. In this paper, we charac-terize the ACCT and related performance metrics for wireless mo-bile networks under a newly proposed general channel allocationscheme. This scheme generalizes the nonprioritized scheme, the re-

served channel scheme, the queueing priority scheme and the sub-rating scheme in such a way as to reduce the blocking probabilityof the handoff calls while keeping the ACCT as long as possible.Explicit formulae for the distribution and the expectation of theACCT are obtained. The call completion probability, the call dropprobability, and the average actual call connection times for boththe complete calls and the incomplete calls are derived. The resultscan form the basis for designing better billing rate schemes by dif-ferentiating incomplete calls and complete calls.

Index TermsActual call connection time, blocking probability,call dropping probability, handoff, mobility, personal communica-tion service, wireless networks.

I. INTRODUCTION

T HE NEXT-generation wireless mobile networks promiseto provide a wide variety of services such as voice, data,and multimedia to users on the move. Mobile users can makea phone call as in wired telephony, or make a connection toretrieve information messages such as email or stock informa-tion, even make a connection to surf the Internet, or to do busi-ness over the Internet (electronic commerce or mobile com-merce). Due to the limited resources (time, frequency, and code)

Manuscript received September 20, 2000; revised August 10, 2001;accepted August 30, 2001. The editor coordinating the review of this paperand approving it for publication is B. Li. The work of W. Li was supported

in part by the Louisiana Board of Regents Support Fund under ContractLEQSF(2000-03)-RD-A-40. The work of Y. Fang was supported in part by theNational Science Foundation Faculty Early Career Award under Grant ANI0093241 and in part by the Office of Naval Research Young Investigator Awardunder Grant N000140210464.

W. Li waswith the Department of Electricaland Computer Engineering, Uni-versity of Louisiana at Lafayette, Lafayette, LA 70504-3890 USA. He is nowwith the Department of Electrical Engineering and Computer Science, Univer-sity of Toledo, Toledo, OH 436063390 USA (e-mail: [email protected]).

Y. Fang is with the Department of Electrical and Computer Engineering, Uni-versity of Florida, Gainesville, FL 32611 USA (e-mail: [email protected]).

R. R. Henry is with the Department of Electrical and Computer En-gineering, University of Louisiana, Lafayette, LA 70504, USA (e-mail:[email protected]).

Digital Object Identifier 10.1109/TWC.2002.804191

of such networks, ongoing call connections may be disrupted ordropped, leading detrimental effect on customer satisfaction. Inorder to address this issue, it seems a good idea to apply differentcharging rates to the complete call connections and incompletecall connections([8], [17]). Thus, the actual call connection time(ACCT), the time a mobile user actually uses the network ser-vices during a call connection, and the ACCT for both incom-plete call connections and complete call connections will be im-portant quantities for such design of billing schemes.

Wireless networks such as personal communication service(PCS) networks have the service coverage area populated withbase stations, which are interconnected with either public tele-phone networks or packet switching networks, the backbonenetworks for providing services to mobile users. Each base sta-tion is responsible forrelaying service to mobileusersin itsarea,called a cell. In each cell, there are usually two classes of calltraffic: new calls and handoff calls. A new call is the call con-nection which initiates in the current cell, while a handoff call isthe call connection which initiates in another cell and is handedover to the current cell. When a new call originates in the cur-rent cell, if an unused channel is available, it will be assigned forthe communication between the mobile user and the base stationand a new call is accepted for service. If a channel cannot be as-signed to the new call, based on the channel allocation scheme,the new call may be blocked and cleared from the system, inwhich case, the ACCT for this call is, therefore, zero. When anew call gains a channel, it will keep the channel either until itis completed in the cell or until the mobile moves out of the cell(we do not consider the channel reassignment in this paper). Ifthe call is completed in the cell, the ACCT is equal to the re-quested call connection time (RCCT). If the mobile user doesnot finish the service in the originating cell, it will move out ofthe cell and attempt to gain service (a channel) in another cell. Ifthe call does not acquire resource (either buffers or a channel),it will be forced to terminate. In this case, the ACCT will be thetime from the start of the call to the time that the call is termi-

nated prematurely. If the handoff is successful, then the call willcontinue in the new cell, and the ACCT will accumulate and thesame procedure repeats.

As we observe from the above discussion, the ACCT dependson the channel (resource) allocation scheme used for handlingnew calls and handoff calls. The following is a brief review ofchannel allocation schemes found in the current literature.

Nonprioritized Scheme (NPS): New calls and handoff callsare treated equally. Thus, either a new call or a handoff callwill be served as long as there is a channel available in thecell. This scheme has been employed by the typical radio

1536-1276/02$17.00 2002 IEEE
http://-/?-http://-/?-http://-/?-http://-/?-


2/10

LIet al.: ACTUAL CALL CONNECTION TIME CHARACTERIZATION FOR WIRELESS MOBILE NETWORKS 683

Fig. 1. Proposed GCAS state machine description.

technologies proposed for PCS systems operated around2 GHz [20].

Reserved Channel Scheme (RCS) (also called guardchannel or cutoff priority scheme): In this scheme,

out of a total of channels in eachcell are reserved for handoff calls. Thus, a new call cangain a free channel only if at least channels in thecell are free. A handoff call will always gain a channelas long as there is any channel available. Notice thatonly the number of channels, not individual channels, arereserved. The overall effect of this scheme is that new

calls will be blocked if at least channels are inuse, while handoff calls will always be served if there areidle channels in the cell [18], [23].

Queueing Priority Scheme (QPS): New calls are blockedwhen there are no free channels available in the cell. Whenthe handoff calls find that all channels are occupied, theyare queued according to various queueing disciplines andwait for an occupied channel to be released [20], [16].

Subrating Scheme (SRS): In this scheme, certain channelsare allowed to be temporarily divided into two channelsat half of the original rate to accommodate handoff calls.This SRS will be activated when all channels are occupiedat the moment of a handover call arrival. When a subrated

channel is released, it forms into an original full-ratedchannel by combining with another subrated channel [17],[21].

In this paper, we propose the following more general channelallocation scheme, which combines the idea of a guard channel,queueing and SRSs, which is described as follows.

1) Each cell consists of channels of rate kb/s and a bufferwith size of for handoff calls only. ( )channels in each cell are reserved for handoff calls andeach of them can be split into two channels with differentrates kb/s and kb/s, respectively, when a handoff callarrives and finds ( ) calls in the cell.

2) A new call will gain a full rate channel for service when

it arrives and finds there are only ( ) callsin the cell. Otherwise, the new call will be blocked andcleared from the system.

3) A handoff call will also gain a full rate channel for servicewhen it arrives and finds the total number of calls in thecell is less than . However, if a handoff call finds allchannels are busy upon its arrival and the number of splitchannels in the cell is less than , one of the full ratechannel will be split into two split channels, one keepingthe original call and the other one being assigned for thecoming handoff call. If the number of split channels isupon the handoff calls arrival, the call will be placed onhold until the time at which either it gains a channel for

service, or it disconnects from the system because of theusers impatience, or the mobile moves out of the currentcell. If a handoff call finds all positions in the handoffqueue are occupied when it comes, it will be forced toterminate and is cleared from the system.

A state machine description of this proposed general channelallocation scheme is given in Fig. 1.

The purpose of our proposed scheme is to incorporate thecommonly used schemes of NPS, RCS, QPS, and SRS in such away as to cast the channel allocation scheme in a unified frame-work, thus, we can reduce the blocking probability of handoffcalls while keeping the ACCT as long as possible. Our focusof this paper is to find an unifying expression of the averageACCT for a call connection, for a complete call connection andfor an incomplete call connection, respectively, under the gen-eral channel allocation scheme. Based on the average ACCT fora complete call and an incomplete call, the operating companiescan design more reasonable charging scheme as already shownin some of the co-authors work [8], [18]: incomplete calls arecharged less for better customer care. Operating companies can,based on the ACCT for complete calls and incomplete calls, de-sign a scheme to minimize the overall revenue loss while takingcare of the customers satisfaction by charging complete callshigher rate and incomplete call lower rate. Moreover, it is intu-itive that the blocked handoff calls and the incomplete calls cansignificantly be reduced because the general channel allocationscheme (GCAS) possesses all the advantages for the handoffcalls in the four schemes. In fact, this is the main advantage ofour scheme if one compares it with other four schemes. Fur-thermore, the ACCT of the new scheme tends to be longer thananyone obtained based on the four schemes because of morechances for handoff calls to gain access, due to incorporationof queueing and subrating ideas in our new scheme. In fact, thenew scheme are expected to include all advantages of the fourschemes from viewpoint of billing strategies and the minimiza-tion of the blocked handoff calls. One disadvantage of this gen-eral scheme is that some interested results, such as ACCT for

any call, the ACCT for a complete call and for an incompletecall, may be too difficult to achieve analytically under more gen-eral assumptions on the cell traffic and holding time distribu-tions due to the complicated operational structure of the channelallocation scheme. This paper, however, is taking the first step toobtain the closed form results for the ACCT for a call, the ACCTfor an incomplete calland a complete call. Itis our beliefthatourresults can be used to develop enhanced billing schemes, whichwill provide for increased satisfaction for the network serviceprovider and the customers.

This paper is organized as follows. In Section II, we listassumptions and notations used in this paper. We then presentanalytical characterization of the ACCT in Section III. The
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


3/10


analysis for the ACCT for incomplete calls and complete callsare given in Section IV. In Section V, we present results forthe cases when the call processes are independent call arrivalprocesses, dependent call arrival processes, and the correlatedcall arrival processes, respectively. We conclude the paper inSection VI.

II. ASSUMPTION ANDNOTATION

In this section, we provide a detailed description for GCAS

and notation we use in this paper. Relationships between this

scheme and the previous schemes are also discussed. The basic

assumption of the GCAS, except those already described in

Section I, are as follows.

1) The wireless mobile network here is assumed to be ho-

mogeneous (see Remark 1 and paper [25]), i.e., the un-

derlying processes and parameters for all cells within the

cellular networks are statistically identical.

2) Each cell consists of channels and a buffer of size

for handoff calls only. ( ) channels in each

cell can be split into two channels with rates -kb/s and

-kb/s, respectively, when a handoff call arrives and find

( ) calls in the cell. The original rate

for each channel is -kb/s, which is the sum of -kb/s and

-kb/s.

3) Each handoff call arriving at the cell and having to wait in

the handoff area has a patience time. If the virtual waiting

time (i.e., the time which the handoff call would have to

wait until service) exceeds the patience time, the handoff

call departs from thesystem andis lost by impatience. The

patience times of the handoff calls are assumed to be inde-

pendent and identically distributed (i.i.d.) with exponen-

tial distribution and with an expectation . See [1][4],[12], and [27] for more application of the impatience phe-

nomena in various systems.

4) The RCCT, , of a call is defined as the call duration

without forced termination. This time is also called the

call holding time([6]) or theunencumbered session time

([23]). As usually [14], [21], it is assumed to be exponen-

tially distributed with mean on the original channel

with rate -kb/sand exponentially distributed with mean

( ,2) on the split channels with rate -kb/s,

where .

5) The cell residence time, , of each mobile is defined as

the interval that a mobile user stays in a cell and is as-

sumed to be exponential distribution with rate . Usually,it depends on the mobile velocity and direction as well as

the shape of the cells. Some reasons for assumptions of

the exponential distribution for RCCT and cell residence

time are listed in Remark 2.

6) Handoff area has been defined as the area in which the

average received power level from the base station is

between the handoff threshold level and the receiver

threshold level [13]. Here, the handoff threshold is set

at the point where the power received from the BS of

a neighboring cell site has started to exceed the power

received from the current BS by a certain amount and/or

for a certain time and the receiver threshold is the point at

which the received power from the BS is at the minimum

acceptable level [26]. The handoff area residence time

is a random variable and usually also depends on the

mobile velocity and direction as well as the shape of the

cells. We use the mean of this variable to approximately

produce an exponential time with same mean, which

is denoted by . This time has the relationship with

the queueing time. In fact, based on our scheme, thequeueing time is the minimum of this variable and the

patient time.

7) All random variables above are assumed to be indepen-

dent each other.

It is worthwhile to point out that if the arrival processes of

the new calls and handoff calls are independent each other and

are Poisson processes, our general scheme will reduce to some

popular schemes as follows.

If and , the scheme is the same as the NPS.

If , , and , the scheme

is equivalent to the RCS, in which the total number of

channels in a cell is and number of reservedchannels for handoff calls is .

If , the scheme is the same as the QPS.

If and , the scheme becomes the

SRS.

Remark 1: The homogeneity assumption implies that the

structures of all cells within the PCS networks are statistically

identical. This is possible in the reality if the network areas are

almost similar, for example, if there are no high mountains,

no large lakes. Most performance studies in cellular networks

commonly invoke this assumption in order to obtain tractable

analytical results. This is particularly true for PCS networks be-

cause the cell size is reduced significantly. The main technicaldifficulty for the nonhomogeneous case is the lack of realistic

modeling of the interactions of the call traffic among cells

(particularly the neighboring cells). To the authors knowledge,

there is no good model for nonhomogeneous cellular networks

in the current literature.

Remark 2: Since we are only handling call connection level

performance, the requested call connection time corresponds to

the regular call holding time in wired telephone systems for

voice services and session time for data services in computer

systems. The exponential model has been shown to appropriate

for call holding time in telephone systems in the past. In [ 24],

the authors also observed that some TCP session time, such as

for remote login, is well modeled by exponential distribution.Moreover, we use the exponential assumption mostly because

we want to derive analytical results as commonly done in the

current literature. The focus of this paper is not on whether these

assumptions are appropriate or not. However, for more general

discussions on the validity of such assumptions, please refer to

some papers by one of the coauthors (see [6], [7], and references

therein). Under NPS and RCS schemes, the exponential distri-

bution has been extended to the more general model (such as

hyper-Erlang) for call performance evaluation ([5], [9], [18]).

Remark 3: It is also noteworthy to point out that there is no

specific assumption on the call arrival processes in the GCAS.

Potential processes such as independent call arrival processes,
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


4/10
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


5/10


traverses the th cell during its call connection. Due to the i.i.d.

assumption, ( ) are i.i.d. random variables. Let

denote its generic form. By a direct probability argument, we

have

(3.4)

(3.5)

and

(3.6)

where , ( ) is the actual waiting

time of a handoff call in the th handoff area during its connec-

tion period, and is the handoff callsmaximum queueing time

in the th handoff area.

Based on the results in (3.3), (3.4), (3.5), and (3.6), we obtain

the distribution function for the ACCT as shown in (3.7) at the

bottom of the page.

We are more interested in the closed expressions of the L-S

transform and also the expectation of the actual call connec-

tion time in the research. For this purpose, we first need to

know some properties of distribution of random waiting time

based on the result in (3.7). We then find the simple closed

expressions for both of these two interested measures.

Based on the assumptions of the network, it is easily to know

that are i .i.d. random v ariables. Let be i ts g eneric

form. By a direct probability argument, we have

arriving handoff call finds

calls in the handoff area

(3.8)

Therefore, the Laplace-Stieltjes transform of the probability dis-

tribution of random variable is given by

(3.9)

and the average actual waiting time of a handoff call in the

handoff area is given by

(3.10)

(3.7)


6/10


Now, from results and (3.7), if we denote by the L-S

transform of probability distribution of the ACCT , we have

(3.11)

From this result, it is readily to know that the expected ACCT

is then given by

(3.12)

Remark 4: From the above simple formulae in (3.11) and

(3.12), we can easily find the related results for the four

channel allocation schemes described in the introduction.

For example, if let in the equation (3.12), we have

. This is the same as that obtained

for RCS in [18] when , by noticing that in this

case could be reasonably deemed as . In fact, as for NPS

and SRS, the formulae of the average ACCT are the same as

that in RCS with a slightly modification of the expressions for

and only based on different schemes. The formula of

the average ACCT for the case of QPS, however, can not be

so simple as those in RCS, NPS and SRS. This formula can

directly be obtained from result in (3.12) by applying .

IV. COMPLETECALLS ANDINCOMPLETECALLS

As we mentioned earlier, it is important to distinguish com-

plete calls from incomplete calls. Due to such consideration, we

are interested in the call completion probability, the call drop-

ping probability, and the actual call connections times for com-

plete calls and incomplete calls.

A. Probabilities

Based on the relationship between the actual call connection

time and the requested call connection time of a new call, we can

classify the actual call connection times into three categories.

The first category occurs when the new call has failed to

be connected. This means that the actual call connection

time is zero.

The second category occurs when the new call is con-

nected and completed. This implies that:

the call completes all necessary handoffs successfully;

the ACCT is greater than zero and is also greater than

or equal to the RCCT.

The final category occurs when the new call is connected

but is eventually forced to terminate. This implies that the

ACCT is greater than zero, and that:

the ACCT is greater than or equal to the RCCT, but the

call is eventually forced to terminate during a handoff,

or

the ACCT is less than the RCCT.

Based on above observation and the results in equations (3.4),(3.5), and (3.6), we have the following results.

The probability that a call is complete (the call completion

probability) is

(4.13)

For incomplete calls, we have the following two situations.

1) The probability that a call is incomplete and the ac-tual call connection time is greater than or equal to

the requested call holding time is given by

(4.14)

2) The probability that a call is incomplete and the ac-

tual call connection time is less than the requested

call holding time is given by

(4.15)
http://-/?-http://-/?-


7/10


Therefore, the probability that a call is incomplete is

(4.16)

It is easy to see that . This is consistent

with the intuition that a new call may be blocked in the initial

cell, or may be connected with the initial cell and ends with anincomplete call or a complete call. It is also noted that our results

are the same as those in [20] for a specific scheme.

B. Average ACCTs for Complete Calls and Incomplete Calls

The average actual call connection times for complete calls

and incomplete calls are of great interest. We determine analyt-

ical results for them in this section.

Let denote the indicator function of the event , that is,

is equal to one if occurs and zero otherwise. Let

and denote the actual call connection times for a complete

call and an incomplete call, respectively. From (3.4), the average

ACCT for a complete call is given by

(4.17)

From (3.5) and (3.6), we obtain the average ACCT for an

incomplete call

(4.18)

Notice that the result of average ACCT in (3.12) can also be

obtained by substituting (4.17) and (4.18) into the expression

.

V. COMPUTATION OFPROBABILITIES AND

We observe that in order to apply our results, we need to find

the new call blocking probability and the forced termina-

tion probability of a handoff call . According to the results in

Section III, we know that these two probabilities are closely de-

pendent on the steady state probabilities of number of thecalls in

the cell when a handoff call and an new call arrive to the cell, re-

spectively. Now, let denote the steady state probability when

there are calls in the cell. It is straightforward to observe that

in general if the handoff arrival process is not Poisson

process and that in general if the new arrival process

is not Poisson process. To show this difference and to finally

find the probabilities and , in this section, we present the

results for cases when the call arrival processes are modeled by

the independent call Poisson arrival processes, the dependent

call arrival processes and the correlated call arrival processes,

respectively.

A. Independent Call Arrival Processes

In this case, the new calls and handoff calls are assumed asPoisson arrival processes and are independent of each other. Ac-

cording to the well-known Poisson arrivals see time averages

(PASTA) result in queueing theory, we know that in this case

for all possible . Now, suppose that the new

call arrival process is Poisson with rate , and that the handoff

call arrival process is Poisson with rate . By a direct proba-

bilistic analysis (Markov chain theory), we have the equation at

the bottom of the page, where and the first

equation at the bottom of the next page.

Therefore, from the results in (3.2) and (3.1), the new call

blocking probability is given by

and the probability that a handoff call is forced to terminate in

a handoff area is given by

B. Dependent Call Arrival Processes

In some wireless networks, call arrival processes may be de-

pendent on the number of call connections in a cell [ 4], [19].

In this case, and will be different from . Suppose that

if

if

if
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


8/10


the new call arrival process is Poisson with rate and that the

handoff call arrival process is Poisson with rate when there

are calls in the cell. Using a similar argument as that in [19],

we obtain the second equation at the bottom of the page, where

Now, we consider the average number of handoff calls over a

longer period of time, say , when the system is in equilibrium.

We observe that the average number of handoff call arrivals is

and that the average number of handoff calls

finding calls in the current cell is . Thus, the probability

that an arriving handoff call finds that there are calls in the cell

is

Similarly, the probability that a new call finds that there are

calls in the cell is

Therefore, from the results in (3.1) and (3.2), we obtain the

new call blocking probability is

and the probability that a handoff call is forced to terminate in

a handoff area is

C. Correlated Call Arrival Processes

In this situation, we consider the MAP, which contains both

phase type renewal processes and correlated arrival streams,

such as theMarkov modulated Poisson process (MMPP), as spe-

cial cases. For the application of the MAP in the PCS networks,

please refer to [17] and the references therein. More precisely,

let be the irreducible infinitesimal generator of an state

Markov process. The sojourn time in state is exponentially dis-

tributed with parameter , . At the end of the

sojourn time in state , three kinds of transitions may take place:

if

if

if
http://-/?-http://-/?-http://-/?-http://-/?-


9/10
http://-/?-http://-/?-


10/10


[11] D. P. Gaver, P. A. Jacobs, and G. Latouche, Finite birth-and-deathmodels in randomly changing environments, Adv. Appl. Probability,vol. 16, pp. 715731, 1984.

[12] R. B. Haugen and E. Skogan, Queueing systems with stochastic timeout,IEEE Trans. Commun., vol. COM28, pp. 19841989, 1980.

[13] D. Hong and S. S. Rappaport, Traffic model and performance analysisfor cellular mobile radio telephone systems with prioritized and nonpri-oritized handoff procedures,IEEE Trans. Veh. Technol., vol. VT-35,pp.7792, 1986.

[14] M. Ivanovich, M. Zukerman,P.Fitzpatrick, and M. Githts, Performancebetween circuit allocation schemes for half- and full-rate connections inGSM,IEEE Trans. Veh. Technol., vol. 47, pp. 790797, Aug. 1998.

[15] M. Kawarazaki, H. Saito, and H. Yamada, An analysis of statisticalmultiplexing in an ATM transport network, IEEE J. Select. AreasCommun., vol. 9, pp. 359367, Apr. 1991.

[16] M. D. Kulavaratharasah and A. H. Aghvami, Teletraffic performanceevaluation of microcellular personal communication networks (PCNs)with prioritized handoff procedures, IEEE Trans. Veh. Technol., vol.48, pp. 137152, Jan. 1999.

[17] W. Li and A. S. Alfa, APCSnetwork with correlated arrival processand splitted-rate channels,IEEE J. Select. Areas Commun., vol. 17, pp.13181325, July 1999.

[18] , Channel reservation for hand-off calls in a PCS network,IEEETrans. Veh. Technol., vol. 49, pp. 95104, Jan. 2000.

[19] W. Li, K. Makki, and N. Pissinou, Performance analysis of a PCS net-work with state-dependent calls arrival processes and impatient calls,

Comput. Commun., vol. 25, pp. 507515, 2002.[20] Y. B. Lin, S. Mohan, and A. Noerpel, Queueing priority channel as-signment strategies for PCS handoff and initial access, IEEE Trans.Veh. Technol., vol. 43, pp. 704712, Aug. 1994.

[21] Y. B. Lin, R. Anthony, and D. J. Harasty, The subrate channel assign-ment strategy for PCS hand-offs, IEEE Trans. Veh. Technol., vol. 45,pp. 122130, Feb. 1996.

[22] K. Sriram and W. Whitt, Characterizing superposition arrival processesin an ATM transport network, IEEE J. Select. Areas Commun., vol. 4,pp. 359367, 1986.

[23] P. V. Orlik andS. S. Rappaport, A model forteletrafficperformanceandchannel holding time characterization in wireless cellular communica-tion with general session and dwell time distributions, IEEE J. Select.

Areas Commun., vol. 16, pp. 788803, June 1998.[24] V. Paxon and S. Floyd, Wide area traffic: The failure of poisson mod-

eling, IEEE/ACM Trans. Networking, vol. 3, pp. 226244, June 1995.[25] S. S. Rappaport, The multiple-call hand-off problem in high-capacity

cellular communications systems, IEEE Trans. Veh. Technol., vol. 40,pp. 546557, 1991.[26] S. Tekinary andB. Jabbari, Handover and channel assignmentin mobile

cellular networks,IEEE Commun. Mag., pp. 4246, Nov. 1991.[27] Y. Q. Zhao and S. A. Alfa, Performance analysis of a telephone system

with both patient and impatient customers, Telecommun. Syst., vol. 4,pp. 201215, 1995.

Wei Li (M99) received the B.Sc. degree fromShannxi Normal University, Xian, China, in 1982,the M.Sc. degree from Hebei University of Tech-nology, Tianjin, China, in 1987, and the Ph.D. degreefrom the Chinese Academy of Sciences, Beijing,China, in 1994.

He is an Associate Professor in the Department ofElectrical Engineering and Computer Science, Uni-

versity of Toledo, Toledo, OH. From 2000 to 2002,he was named as the Bell South/BORSF Professor inTelecommunications. From 1999 to 2002, he was an

Assistant Professor in the Department of Electrical and Computer Engineering,University of Louisiana, Lafayette. He engaged in research and teaching at theUniversity of Manitoba, Winnipeg, Canada, and the University of Winnipeg,Winnipeg, Canada, from 1995 to 1999. He was once a Senior Researcher atthe University of Newcastle, England and an Associate Researcher at the Chi-nese Academy of Sciences. His current research interests are in the evaluation,design and implementation of teletraffic models for personal communicationsservices networks, wireless networks, and mobile satellite communications net-works as well as the applications in these areas of operations research, queueingnetworks, and reliability networks.

Yuguang Fang (S92M94SM99) received theB.S. and M.S. degrees in mathematics from QufuNormal University, Qufu, Shandong, China, in 1984and 1987, respectively, the Ph.D degree in systemsand control engineering from the Department ofSystems, Control and Industrial Engineering, CaseWestern Reserve University, Cleveland, OH, inJanuary 1994, and the Ph.D degree in electricalengineering from the Department of Electrical and

Computer Engineering, Boston University, Boston,MA, in May 1997.From 1987 to 1988, he held research and teaching positions in both the De-

partment of Mathematics and the Institute of Automation at Qufu Normal Uni-versity. From September 1989 to December 1993, he was a Teaching/ResearchAssistant in the Department of Systems, Control and Industrial Engineering,Case Western Reserve University, where he held a Research Associate positionfrom January 1994 to May1994.FromJune 1994to August1995, heheld a Post-doctoral position in the Department of Electrical and Computer Engineering,Boston University. From September 1995 to May 1997, he was a Research As-sistant in the Department of Electrical and Computer Engineering, Boston Uni-versity. From June 1997 toJuly 1998,he wasa VisitingAssistant Professorin theDepartment of Electrical Engineering, University of Texas, Dallas. From July1998 to May2000, he wasan Assistant Professorin theDepartment of Electricaland Computer Engineering, New Jersey Institute of Technology, Newark. SinceMay 2000, he has been an Assistant Professor in the Department of Electricaland Computer Engineering, University of Florida, Gainesville, FL. His research

interests span many areas including wireless networks, mobile computing, mo-bile communications, automatic control, and neural networks. He has publishedover 80 papersin refereed professionaljournalsand conferences.He is an Editorfor ACM Wireless Networks, an Area Editor for ACM Mobile Computing andCommunications Review, an Associate Editor for Wiley International Journalon Wireless Communications and Mobile Computing. He is also actively in-volved with many professional conferences such as ACM MobiCom01, IEEEINFOCOM98, INFOCOM00, IEEE WCNC02, WCNC00 (Technical Pro-gram Vice-Chair), WCNC99, and the International Conference on ComputerCommunications and Networking (IC3N98) (Technical Program Vice-Chair).

Dr. Fang has also engaged in many professional activities. He is a Member ofACM. He is an Editor for IEEE TRANSACTIONS ONCOMMUNICATIONS, an Ed-itor for IEEE JOURNAL ONWIRELESSCOMMUNICATIONS. He is Feature Editorfor Scanning the Literature in IEEE PERSONALCOMMUNICATIONS. He has re-ceived the prestigious National Science Foundation Faculty Early CareerAwardin 2001 and the Office of Naval Research Young Investigator Award in 2002.He is listed in MarquisWhos Who in Science and Engineering,Whos Who in

America, andWhos Who in World.

Robert R. Henry (M67SM83) received theB.S.E.E. and M.S.E.E. degrees from the Universityof Southwestern Louisiana, Lafayette, in 1967 and1969, respectively, and the Ph.D. degree in electricalengineering from New Mexico State University, LasCruces, in 1974.

He is currently a Professor and Department Headof the Electrical and Computer Engineering Depart-ment, University of Louisiana, where he has been aFaculty Member for 25 years. Previously, he servedtwo years on the faculty in the Electrical Engineering

Department, Northern Arizona University, Flagstaff. He has published and pre-sented over 60 papers and articles in the computer control and communica-tions systems areas. He was a Senior Engineer for the Philco Ford Corporation,Houston, TX, where he conducted telecommunications research for NASA atthe Johnson Space Center, Houston, TX. He has presented several industrialshort courses and serves as a consultant for the telecommunications industry.He is recognized as a forensic engineering expert.

Dr. Henry is a registered Professional Engineer in the state of Louisiana.

Documents

Actual Connection Time